Determinants of director compensation in two-tier systems: evidence from German panel data

From now on, we will use the phrase director for any non-executive board member and executive or manager for all executive directors.

It is well known that the German corporate governance system is characterized by a two-tier system with two boards: the supervisory board (Aufsichtsrat) and the management board (Vorstand). According to the German Stock Corporation Act (Aktiengesetz—AktG), the supervisory board supervises (§111 AktG) and appoints (§84 AktG) members of the management board. A peculiarity of the German corporate governance system is the Codetermination Act of 1976 (MitbestG). It regulates the possibility of mandatory employee representatives within the supervisory board depending on firm size and the sector the firm is operating in. Thus, for firms operating under codetermination there are two types of supervisory board members: shareholder representatives and employee representatives. As both director types are subject of the same rights and duties, we do not explicitly distinguish between their backgrounds in the remainder of the paper. Nevertheless, the sensitivity to monetary incentives will be generally lower for employee representatives since they are usually obliged to pass a substantial share of their compensation package on to Hans-Böckler-Stiftung, a foundation of the Confederation of German Trade Unions. However, this does not mean that compensation is irrelevant for employee representatives: First, the general provisions with regard to the payout ratio to employee representatives have been gradually revised in the past years (higher thresholds until a mandatory and proportionate transfer payment kicks in and a removal of total compensation caps for these directors). Second, the remaining residual board compensation will be still a sizeable portion of the total compensation package of employee representatives. Therefore, the quest for optimal compensation is also applicable to boards under codetermination, apart of the fact that shareholder representatives will have a casting vote in the event of deadlocks anyway.

With respect to the result of this bargaining process, two competing views emerged: the optimal-contracting and the managerial-power perspective (see Grossman and Hart 1983; Bebchuk et al. 2002 for an overview of the arguments). While the optimal contracting view presumes that shareholders establish compensation policies, i.e. the bargaining power is allocated to shareholders but agents will take advantage of information asymmetries, in the managerial power framework agents are able to set their own pay (Ruiz-Verdú 2008). Concerning executive compensation, recent evidence suggests that both perspectives are needed to fully understand existing practices (e.g. Bruce et al. 2005; Schmidt and Schwalbach 2007; Dittmann and Maug 2007; Fahlenbrach 2009).

To illustrate, tournament theory has been used in executive compensation studies to provide a theoretical justification for pay differentials among the top management team and its effect on corporate performance (see e.g. O′Reilly et al. 1988; Conyon et al. 2001). However, contrary to executive compensation, director compensation is set for a group of individuals (i.e. the board) as a whole (§ 113 AktG). Interdirectoral differences in compensation stem from assuming additional functions (e.g. chairman or committee membership) or variations in meeting attendance (Farrell et al. 2008).

The keywords are “board of directors compensation”, “director incentive pay”, “outside director compensation”, “board remuneration” and “board incentive contracts”. The shortlist of key journals includes (amongst others) the Journal of Finance, the Journal of Financial Economics, the Review of Financial Studies and the Journal of Corporate Finance.

Hence, our review does not consider working papers that deal with director compensation determinants. However, we will refer to them in our development of the hypotheses and the discussion of the results. Relevant papers that we have identified via SSRN are from Perry (1999), Bryan et al. (2000), Bryan and Klein (2004) and Adams (2003). Additionally, we searched the database of the Deutsche Nationalbibliothek for relevant dissertations that have not found their way into refereed journals. We could not find any dissertations that met our requirements.

Various changes in German law strengthened the position of the board and professionalized its service, e.g. Corporation Control and Transparency Act (KonTrAG) 1998 or the introduction of the German Corporate Governance Code (DCGK) 2002. See Nowak (2001), Theisen (2003a) and Goergen et al. (2008) for an overview of the legislative efforts and their implications.

Given the wide adoption of stock-based incentive programs in the US, this evidence has been observed to be more consistent based on capital market-related (e.g. total shareholder return) rather than accounting-based indicators (e.g. return on assets). However, in contrast to US findings, we expect director compensation in Germany to be more closely related to accounting-related performance measures compared to capital market performance. Moreover, we expect fewer stock-based incentives for directors in German companies, since the German Federal Court of Justice prohibited the use of stock options as a component of director pay in the 2004 “Mobilcom” decision (BGH II ZR 316/02). As a consequence, only contractual replications such as phantom stocks or stock appreciation rights are permitted.

Other researchers argue that concentrated ownership represents a prerequisite that shareholders are able to establish efficient contracts and thus suggest a complementary relationship (e.g. Hartzell and Starks 2003). A positive correlation between compensation level and governance mechanisms is, however, also in line with a managerial/director power view. See Fahlenbrach 2009 for a thorough discussion of the empirical predictions of these views for the case of executive compensation.

This is also in line with recent evidence concerning compensation of German executives (Kaserer and Wagner 2004; Rapp and Wolff 2010).

Note that due to German regulation, e.g. the Codetermination Act, supervisory boards of German firms are about twice the size compared to the proportion of non-executive directors in boards of a one-tier system (Baums 2001).

While the agency-oriented finance literature is rather sceptical concerning busy directors, there is a large body of management literature that highlights the positive spillover effects of these networks. See Hillman et al. (2008) and Oehmichen et al. (2009) for a discussion.

Consequently, commentators argue that the board should be dominated by firm-external directors who are expected to be less susceptible to collusion and being captured by powerful CEOs (Lipton and Lorsch 1992; Jensen 1993). This idea has also made its way into German policy reforms: according to a recent amendment to the German Stock Corporation Act, former members of the executive management must await a two-year cooling-off period before they can assume a position on the supervisory board. However, the shareholders' meeting may opt out if the combined stake of the nominating shareholders exceeds 25% (§100 AktG). Additionally, the German Corporate Governance Code recommends a maximum of two seats for former executives in supervisory boards in total. However, some academics raise concerns that boards dominated by outside directors may face negative effects: former executives are supposed to have valuable firm-specific knowledge that reduces information costs (e.g. Donaldson and Davis 1994).

This condition was introduced because we want the samples to be exactly the same when comparing coefficients across regression models. Otherwise, the computation of likelihood-ratio tests comparing models cannot be completed and any interpretations of why the results have changed must take into account differences between the samples. In addition, we decided to track stock volatility over a 36-month period requiring at least 3 years of data preceding any sample year.

We are aware of the fact that this approach does not differentiate between different board members (e.g. chairman and ordinary directors) and may introduce a bias into our compensation data if board composition changes during the financial year. However, we chose this approach as it would take enormous effort to compile the relevant data at director level. Additionally, only roughly two-thirds of the companies disclose information about individual pay levels, which would in turn further restrict our sample. Finally, by following this procedure we connect to all prior German studies which also applied this calculation (Schmid 1997; Knoll et al. 1997; Elston and Goldberg 2003).

While in the US context commentators are concerned about the role of the CEO, in particular in the case of CEO duality meaning that the CEO is simultaneously chairman of the company, for the two-tier setting, and Germany in particular, it is rather the role of the chairman of the supervisory board that is controversial. One of the reasons for this is the dominance of the chairman, which constitutes itself in the privilege of having a casting vote if elections between shareholder and employee representatives are tied.

Equation 2 describes a general one-way fixed-effects model where α_i is the coefficient of a firm-specific dummy variable. In a random-effects model, the firm-specific effect is presumed to be random with α _i ~ (α, σ _α ² ) and ε ~ (0, σ _u ² ). A standard extension of the individual effects model is a two-way effects specification that allows time effects to be incorporated as well. Then, y _it  = α _i  + γ _t  + x′ _it β + ɛ _it , where γ _t are time-variant (but firm-invariant) effects. Since it is common to assume that the time effects are fixed effects, i.e. non-random, the latter equation reduces to Eq. 2, where the regressors include time dummies. We include time dummies in all our regression models.

We use a Breusch-Pagan LM test to examine whether allowing for firm level heterogeneity improves the fit of our compensation level models (Breusch and Pagan 1980). The null hypothesis that the variance of the group-specific errors component α _i of groups is zero is rejected at ρ < 0.001, suggesting the use of a firm effects model (see Table 5).

Another way of discriminating between random and fixed effects is to define the target of inference (Wooldridge 2002). A random-effects model is more appropriate if the interest of inference relates to a population mean, i.e. units are viewed as sampled from an overall population. In contrast, fixed effects are more suitable if the interest concerns the particular units in the dataset at hand.

The standard Hausman test that is built in most statistical packages requires the random-effects estimator to be efficient, which in turn requires that α _i and ε _it are i.i.d. (Cameron and Trivedi 2005). However, this is an invalid assumption if cluster-robust standard errors differ substantially from ordinary standard errors, which is usually the case in panel estimations (Petersen 2008). Here, a robust version of the Hausman test is needed (Wooldridge 2002). In our analysis, we use a panel-robust version that is described by Arellano (1993) and Wooldridge (2002, 290–291).

For example, in short panels like ours where N > T, consistent estimation of firm-fixed effects by simply including individual intercepts is not possible due to the incidental parameters problem (Cameron and Trivedi 2005). An alternative would be the specification of a conditional logit model (Baltagi 1995; Greene 2005). However, as a consequence all companies that exhibit no variation in the dependent variable over time would be dropped. As this is the case in the vast majority of all companies examined in this study (~90%), application of the conditional logit model would induce major inefficiency and is hence less favorable. Similarly, specifying the intercept as a random variable places heavy restrictions on the data such as low intra-class correlation and only a limited numbers of clusters in order to achieve stable results during quadrature approximation (Lesaffre and Spiessens 2001; Rabe-Hesketh et al. 2005). With cluster size over 200 and a high intra-class correlation (>90%), our results change considerably (i.e. greater than a relative difference of 10⁻² in coefficient estimates) when we refit random-effects logit models with different numbers of integration points. Thus, although we get similar results to those reported in Sect. 4.2 when specifying a random-effect logit model in the default mode in Stata, we choose to restrict our presentation of results to population-averaged models only.

Andreas et al. (2009) show that only 4% of Prime Standard companies who have adopted short-term performance-based pay for directors tie the incentive plan to stock price development. Similarly, Knoll et al. (1997) do not find a significant relationship in their regression results using stock-based performance indicators.

In fact, there is a controversial discussion in Germany about whether or not compensation packages of directors should be tied to firm performance at all. Opponents regularly argue that reliable performance criteria that board members can actually influence are not available. Moreover, they point to the stewardship theory outlined above and generally question the extrinsic motivation structure of directors (e.g. Fallgatter 2003; Böcking 2004). From their perspective, compensation should, if at all, be exclusively tied to the input dimension of performance, for example meeting attendance. This view is also generally supported by labour unions (Seyboth 2003). However, proponents of performance-based compensation bring forward a series of arguments that illustrate how supervisory boards can increasingly influence the firm's performance, for example via management decisions that must be subject to the board’s approval or by indirectly framing the management's behaviour (e.g. Lutter 2001; Fallgatter and Simons 2003). Additionally, based on formal models of a two-tier setting, others emphasize the function of performance-based compensation to provide for congruent interests, to prevent collusion or to attract and retain scarce talent (e.g. Martens 2000; Hartmann 2003).

In fact, there is recent evidence that professional directors increasingly become a scarce resource (e.g. Engeser 2009; Prange 2009).

This is because the variables’ marginal effect on the dependent variable will vary with the magnitude of change in the variable of interest, the variables’ starting value, and the value of all other model variables (Long and Freese 2006).

However, we see that the confidence interval is smaller near the centre of the data and increases as we move to smaller companies, where our dataset has comparatively few observations.

For example, 84.5% of the chairmen characteristics will remain constant over the sample period.

This is because some observations had to be dropped since some industries perfectly predict (non-)adoption.

The German Commercial Code (Handelsgesetzbuch—HGB) stipulates in Section §289a that stock companies must disclose an annual declaration of corporate governance which should also elaborate on the independence of board members. This requirement has been part of a recent corporate governance reform in Germany (Act to Modernize Accounting Law (BilMoG)).